As I understand things: >From the perspective of the router, it is receiving packets from all over and sending them all over. It is reasonable to assume that data is flowing over many ports concurrently. It is also expected to have packets from diverse endpoints being interleaved over common paths.
But from the perspective of the specific endpoints, there is but a single virtual wire meaning that everything is serial. The unit of data is presented to FTP which breaks it down (fragments) it into small (less than 1500 byte) chunks and builds each packet one at a time. It then sends that packet and waits for an acknowledgement (note 1). Once the packet arrives at the router, The router has to insure that the packet is completely received then examine it to figure out where to send it. I guess it then places the packet in a queue for transmission on the selected port. Unlike the originator, the router does not care if the packet is ever received anywhere or gets corrupted. So, my definition of 'performance' is the elapsed time from packet birth to packet death. Since the path is likely to be carrying interleaved packets from multiple sessions, the transit time could be somewhat variable. Note 1 - it does wait, but optionally asynchronously. That is, packet 2 could be sent before the acknowledgment for packet 1 is received. How many packets that can be in this limbo state is a tuning parameter. I am told that there are a number of tricks the network gadgets use to speed things along, compression, streaming, etc where the time in the network can be reduced. But, I would expect each hop to add a significant delay. As in my over simplified model, a multiple of network speed to include congestion. But perhaps not a full multiple due to those tricks. I could be wrong. -----Original Message----- From: IBM Mainframe Discussion List [mailto:[EMAIL PROTECTED] On Behalf Of Tom Marchant Sent: Wednesday, June 25, 2008 5:42 PM To: [email protected] Subject: Re: Slow FTP transfer from z/OS to Unix On Wed, 25 Jun 2008 14:00:00 -0500, Hal Merritt wrote: >I wonder if there is some repackaging along the way. My model assumes >that a single packet traverses the network unchanged in any way. > >A fixed delay at the appliance works, but I don't understand how a >packet that has to be transmitted twice would take the same amount of >time as one transmitted only once. I'm not a network guy. You may be correct when considering the transmission of a single packet. However, when transferring a large file, consisting of thousands of packets, doesn't the router transmit and receive packets in parallel? In that case, wouldn't you expect that the difference in the total time when there are additional hops to be very small? Perhaps even smaller than the variations that occur due to the amount of network traffic. -- Tom Marchant ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html NOTICE: This electronic mail message and any files transmitted with it are intended exclusively for the individual or entity to which it is addressed. The message, together with any attachment, may contain confidential and/or privileged information. Any unauthorized review, use, printing, saving, copying, disclosure or distribution is strictly prohibited. If you have received this message in error, please immediately advise the sender by reply email and delete all copies. ---------------------------------------------------------------------- For IBM-MAIN subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: GET IBM-MAIN INFO Search the archives at http://bama.ua.edu/archives/ibm-main.html
